Dağıtık Sezim Sistemlerinde Beklenti Kuramına Dayalı Başarım Metrikleri için Optimal Karar Kuralları

Dülek B. (Executive)

TUBITAK Project, 2023 - 2026

  • Project Type: TUBITAK Project
  • Begin Date: February 2023
  • End Date: February 2026

Project Abstract

The problem of selecting one among a finite number of hypotheses that explain the state of a phenomenon of interest based on measurements taken under uncertainty and with respect to a specific performance criterion is the subject of detection theory. In distributed detection systems, the measurements made by different sensors are quantized and then sent to a fusion center. Based on the quantized outputs or the local decisions of the sensors, a final decision regarding the state of the phenomenon of interest is declared in the fusion center. Generally, the decision rules employed at the sensors and that at the fusion center are designed to optimize objective performance metrics like Bayes risk, the average probability of error, or the Neyman-Pearson criterion. However, as it is evident from many applications that involve distributed decision tasks such as crowdsourcing and recommendation systems which rely upon people's choices and ratings, the decisions may need to be taken based on human agents' subjective opinions shaped by behavioral, psychological, and social-economic perceptions. Prospect theory was developed in the field of economy to accurately predict humans' decisions in risky situations involving losses and gains. As a Nobel award-winning theory, it has found widespread applications in diverse fields such as economy, finance, tax policy making, pricing in smart grids, and human-centric communications where it is asserted that quality of experience should be taken into account along with the quality of service metrics. At the heart of prospect theory lies the idea that humans perceive losses and gains by passing them through a value function, and likewise, they perceive probabilities by passing them through a weighting function, which is called the framing effect in the relevant literature.

In this project, behavioral (boundedly rational) versions of the objective performance criteria commonly employed in the distributed detection literature are proposed based on mathematical models used in prospect theory. Our aim is to determine the optimal local and fusion center decision rules for distributed detection systems that include human agents. In this way, it will be possible to determine the optimal decision rules under more realistic models for situations where the utility of a final decision is assessed by humans or when the fusion of heterogeneous data collected from both physical sensors and human agents needs to be performed or in human-centric distributed detection applications like crowdsourcing. In addition to providing a rigorous solution to an open problem in the literature, prospect theory-based distributed detection will be investigated comprehensively by working on several novel research problems. This will contribute significantly to the culmination of graduate students' involvement in the field. Since the behavioral performance criteria depend on the decision rule in a highly nonlinear way and standard approaches are not sufficient, the development of a new mathematical method for the solution of the optimal decision rules will become another aspect of the novelty of this project.